The development of means of definition of the optimum ratio of computational algorithm and the reconfigurable structure
Iryna Anatoliivna Klymenko, Oleh Holovko, Maksym Hilliaka, Yaroslav Mytsyo · Eastern-European Journal of Enterprise Technologies · 2016
Known tools for mapping tasks on a parallel computing structure, developed for fixed architectures or switched computing environment, are based on adaptation of a computing algorithm to caused computing structure and, thus cannot be effectively used to solve tasks of big and super big size in reconfigurable computing systems, which have certain software and hardware limitations. This describes the actuality and the value of the completed research.We described and researched physical characteristic of the FPGA and defined main criteria that affect an efficiency of parallel computations based on the reconfigurable technology of the FPGA, particularly communication delays of the physical level of the FPGA chips. The new method to find an optimum ratio between a computing algorithm and a structure of the reconfigurable computing system of the FPGA is proposed. It allows to propose a new reconfiguration strategy, which differs from known by mutual adaptation of a computing algorithm and a computing environment.We proposed and implemented a library of the functional core for the FPGA to solve tasks of linear algebra and matrix operations, which provides the set of functional blocks with optimum characteristics according to defined performance criteria of a reconfigurable computing space. The developed library allows to effectively vary the computation granularity in terms of reconfigurable computations.The proposed tools and the results of the research allowed to increase the efficiency of the process of task mapping on the computing structure of dynamically reconfigurable computing systems, based on the FPGA to solve tasks of big and super big size with regular reconfigurable computing structures.